Claude Code

Claude Code is Anthropic’s coding agent. In an Archil sandbox it gets a Linux VM of its own, so you can let it work without permission prompts. Its workspace and session history live on an Archil disk, so the work outlives the sandbox, and a new sandbox can pick up the same conversation.

Prerequisites

Export your Archil API key, a region where sandboxes are available (aws-us-east-1, aws-us-west-2, or aws-eu-west-1), and an Anthropic API key, then install the SDK:

export ARCHIL_API_KEY=key-...
export ARCHIL_REGION=aws-us-east-1
export ANTHROPIC_API_KEY=sk-ant-...

npm init -y && npm pkg set type=module
npm install disk tsx

pip install archil

Run Claude Code in a sandbox

This script creates a disk, starts a sandbox with the disk mounted at /workspace, installs Claude Code, and gives it a task in print mode (claude -p), which works until the task is done, prints the answer, and exits. Then it deletes the sandbox; everything the agent made is on the disk.

import * as archil from "disk";

const env = {
  ANTHROPIC_API_KEY: process.env.ANTHROPIC_API_KEY!,
  CLAUDE_CONFIG_DIR: "/workspace/.claude", // keep sessions on the disk
  IS_SANDBOX: "1", // sandboxes run as root; allows --dangerously-skip-permissions
};

const { disk } = await archil.createDisk({ name: "claude-code-workspace" });
const sandbox = await archil.createSandbox({ name: "claude-code", mounts: [{ disk, path: "/workspace" }] });
await sandbox.exec("npm install -g @anthropic-ai/claude-code");

const result = await sandbox.exec(
  'claude -p "Create a hello-world Express app with a test, and run the test. Reply in two sentences." ' +
    "--model claude-sonnet-5 --dangerously-skip-permissions",
  { cwd: "/workspace", env },
);
console.log(result.stdout);

await sandbox.stop();
await sandbox.delete();

import os

import archil

env = {
    "ANTHROPIC_API_KEY": os.environ["ANTHROPIC_API_KEY"],
    "CLAUDE_CONFIG_DIR": "/workspace/.claude",  # keep sessions on the disk
    "IS_SANDBOX": "1",  # sandboxes run as root; allows --dangerously-skip-permissions
}

disk = archil.create_disk(name="claude-code-workspace").disk
sandbox = archil.create_sandbox(
    name="claude-code", mounts=[archil.SandboxMountSpec(disk=disk, path="/workspace")]
)
sandbox.exec("npm install -g @anthropic-ai/claude-code")

result = sandbox.exec(
    'claude -p "Create a hello-world Express app with a test, and run the test. Reply in two sentences." '
    "--model claude-sonnet-5 --dangerously-skip-permissions",
    cwd="/workspace",
    env=env,
)
print(result.stdout)

sandbox.stop()
sandbox.delete()

Run it with npx tsx start.ts or python start.py. The agent’s wording varies from run to run. One run printed:

Created a hello-world Express app at `/workspace/hello-express` (`app.js` serving "Hello, World!" at `/`, plus `server.js` to run it) with a supertest/Jest test in `app.test.js`. Ran `npm test` and the test passed (1/1).

Pick up where you left off

Claude Code’s project and session history are on the disk, so any sandbox that mounts it can continue the conversation. This script starts a new sandbox on the same disk and runs Claude Code with --continue, which resumes the most recent conversation in /workspace.

import * as archil from "disk";

const env = {
  ANTHROPIC_API_KEY: process.env.ANTHROPIC_API_KEY!,
  CLAUDE_CONFIG_DIR: "/workspace/.claude",
  IS_SANDBOX: "1",
};

const [disk] = await archil.listDisks({ name: "claude-code-workspace" });
const sandbox = await archil.createSandbox({ name: "claude-code", mounts: [{ disk, path: "/workspace" }] });
await sandbox.exec("npm install -g @anthropic-ai/claude-code");

const result = await sandbox.exec(
  'claude --continue -p "Add a /health route with a test and run all the tests. Reply in two sentences." ' +
    "--model claude-sonnet-5 --dangerously-skip-permissions",
  { cwd: "/workspace", env },
);
console.log(result.stdout);

await sandbox.stop();
await sandbox.delete();

import os

import archil

env = {
    "ANTHROPIC_API_KEY": os.environ["ANTHROPIC_API_KEY"],
    "CLAUDE_CONFIG_DIR": "/workspace/.claude",
    "IS_SANDBOX": "1",
}

[disk] = archil.list_disks(name="claude-code-workspace")
sandbox = archil.create_sandbox(
    name="claude-code", mounts=[archil.SandboxMountSpec(disk=disk, path="/workspace")]
)
sandbox.exec("npm install -g @anthropic-ai/claude-code")

result = sandbox.exec(
    'claude --continue -p "Add a /health route with a test and run all the tests. Reply in two sentences." '
    "--model claude-sonnet-5 --dangerously-skip-permissions",
    cwd="/workspace",
    env=env,
)
print(result.stdout)

sandbox.stop()
sandbox.delete()

Added a `/health` route returning `{ status: 'ok' }` with a corresponding test in `app.test.js`. All tests pass (2/2).

Going further

Stream output. Pass onOutput: ({ data }) => process.stdout.write(data) to exec (in Python, on_output=lambda out: sys.stdout.buffer.write(out.data)) to watch the agent as it works. Add --output-format stream-json --verbose to get one JSON event per line.
Work interactively. Start claude without -p in a pseudo-terminal with sandbox.run; see Opening interactive shells.

Use Vercel AI Gateway instead of an Anthropic key

Claude Code can send its requests through Vercel AI Gateway. Export AI_GATEWAY_API_KEY instead of ANTHROPIC_API_KEY, and replace the ANTHROPIC_API_KEY line in env with these lines. ANTHROPIC_API_KEY must be an empty string, because Claude Code prefers it over ANTHROPIC_AUTH_TOKEN. On the Developer plan, sandboxes can’t reach the gateway, because it isn’t on the managed allowlist; use your provider key instead.

ANTHROPIC_BASE_URL: "https://ai-gateway.vercel.sh/claude-code",
ANTHROPIC_AUTH_TOKEN: process.env.AI_GATEWAY_API_KEY!,
ANTHROPIC_API_KEY: "",

"ANTHROPIC_BASE_URL": "https://ai-gateway.vercel.sh/claude-code",
"ANTHROPIC_AUTH_TOKEN": os.environ["AI_GATEWAY_API_KEY"],
"ANTHROPIC_API_KEY": "",

Clean up

Each script deletes its sandbox when it finishes, so only the disk is left. Delete it:

import * as archil from "disk";

const [disk] = await archil.listDisks({ name: "claude-code-workspace" });
await disk.delete();

import archil

[disk] = archil.list_disks(name="claude-code-workspace")
disk.delete()

Next steps

OpenCode

Run OpenCode headlessly in a sandbox.

OpenAI Codex

Run the Codex CLI headlessly in a sandbox.

Coding agent in a sandbox

Build your own coding agent with the AI SDK instead.

Persistent sandboxes

Lifecycle, services, network access, and forking in detail.

Agents

Analytics

Training

Apps & databases

Use Archil on other platforms

Prerequisites

Run Claude Code in a sandbox

Pick up where you left off

Going further

Clean up

Next steps

OpenCode

OpenAI Codex

Coding agent in a sandbox

Persistent sandboxes

​Prerequisites

​Run Claude Code in a sandbox

​Pick up where you left off

​Going further

​Clean up

​Next steps

OpenCode

OpenAI Codex

Coding agent in a sandbox

Persistent sandboxes

Prerequisites

Run Claude Code in a sandbox

Pick up where you left off

Going further

Clean up

Next steps